Just for clarification. When I read the initial post, my assumption was that the difference between the parallel and serial results was due to a different ordering of operations and was not necessarily a "real" error. I.e. while the serial and parallel results could be different, they could both be equally valid. Is this the case, or is it in fact a true failure in the parallel implementation?<div>
<br></div><div>- Wes<br><br><div class="gmail_quote">On Mon, Oct 26, 2009 at 2:22 PM, Luis Ibanez <span dir="ltr"><<a href="mailto:luis.ibanez@kitware.com">luis.ibanez@kitware.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Hi Kevin,<br>
<br>
Thanks for the clarification.<br>
<br>
<br>
A) From your description, it looks like the code is not going<br>
to get better by keeping it on your side.<br>
<br>
...and I mean that in a good way... :-)<br>
<br>
So, I will suggest that you go ahead and submit it to the<br>
Insight Journal, and include in the paper a note about your<br>
observations related to the different results obtained between<br>
parallel and serial.<br>
<br>
Once the code is in the Insight Journal, it will be easier for<br>
other developers to look at the code, and make suggestions<br>
for improvements, (assuming that this is something caused<br>
by your code), or help locate a potential bug in ITK.<br>
<br>
<br>
B) Providing and explicit test that compares the parallel and<br>
serial results will be great. Given that we know that there<br>
is a problem there, we may as well expose it clearly so<br>
we start working towards fixing it.<br>
<br>
I would think that a "failing" test make more sense than<br>
a "passing" tests in this case.<br>
<br>
<br>
C) Thanks for filing the bug report:<br>
<div class="im"><a href="http://public.kitware.com/Bug/view.php?id=9766" target="_blank">http://public.kitware.com/Bug/view.php?id=9766</a><br>
<br>
</div> I'll take a look at the problem.<br>
<br>
<br>
Thanks<br>
<br>
<br>
Luis<br>
<br>
<br>
------------------------------------------------------------------<br>
<div><div></div><div class="h5">On Mon, Oct 26, 2009 at 1:27 PM, Kevin H. Hobbs <<a href="mailto:hobbsk@ohiou.edu">hobbsk@ohiou.edu</a>> wrote:<br>
> My comments are below :<br>
><br>
> On Mon, 2009-10-26 at 12:57 -0400, Luis Ibanez wrote:<br>
>> A) You could add the test yourself by<br>
>> following the procedure described in:<br>
>> <a href="http://www.itk.org/Wiki/ITK_Procedure_for_Adding_a_Test" target="_blank">http://www.itk.org/Wiki/ITK_Procedure_for_Adding_a_Test</a><br>
>><br>
>> or<br>
>><br>
>> B) You could send us the test's source code<br>
>> and we could take care of adding it to the<br>
>> toolkit.<br>
>><br>
><br>
> I'm sorry, my original e-mail was not clear enough.<br>
><br>
> I was trying to ask what I should do for an Insight Journal submission<br>
> that I'm working on. ( I now have permission to submit. ) I was not<br>
> asking about what I should do if I found a bug in ITK or IJ.<br>
><br>
> I know that my implementation of MPI Fast Marching will not produce the<br>
> exact same results as the serial algorithm but the MPI version seems to<br>
> work well enough for my requirements.<br>
><br>
> Should I directly compare the results from a serial run to those from a<br>
> parallel run and produce a failing test? This would make it very clear<br>
> what the implementation does not do.<br>
><br>
> Or should I produce a passing test that verifies the difference?<br>
><br>
>><br>
>> Please let us know what is your preference.<br>
>><br>
>><br>
>> BTW: It would seem that filing a bug report<br>
>> on this issue may also be a good idea.<br>
><br>
> As to bugs in IJ and ITK: I already filed bug reports (or I have been<br>
> convinced that what I saw was not a bug) on everything I've noticed<br>
> recently.<br>
><br>
> I filed a feature request for IJ to support MPI.<br>
><br>
> I filed a bug report for ITK<br>
> <a href="http://public.kitware.com/Bug/view.php?id=9766" target="_blank">http://public.kitware.com/Bug/view.php?id=9766</a> about how extracting a<br>
> piece from a reader which has had UpdateOutputInformation called but not<br>
> Update produces incorrect results. This bug report has code attached<br>
> which could become a test.<br>
><br>
</div></div><div><div></div><div class="h5">_____________________________________<br>
Powered by <a href="http://www.kitware.com" target="_blank">www.kitware.com</a><br>
<br>
Visit other Kitware open-source projects at<br>
<a href="http://www.kitware.com/opensource/opensource.html" target="_blank">http://www.kitware.com/opensource/opensource.html</a><br>
<br>
Kitware offers ITK Training Courses, for more information visit:<br>
<a href="http://www.kitware.com/products/protraining.html" target="_blank">http://www.kitware.com/products/protraining.html</a><br>
<br>
Please keep messages on-topic and check the ITK FAQ at:<br>
<a href="http://www.itk.org/Wiki/ITK_FAQ" target="_blank">http://www.itk.org/Wiki/ITK_FAQ</a><br>
<br>
Follow this link to subscribe/unsubscribe:<br>
<a href="http://www.itk.org/mailman/listinfo/insight-users" target="_blank">http://www.itk.org/mailman/listinfo/insight-users</a><br>
</div></div></blockquote></div><br><br clear="all"><br>-- <br>Wesley D. Turner, Ph.D.<br>Kitware, Inc.<br>Technical Leader<br>28 Corporate Drive<br>Clifton Park, NY 12065-8662<br>Phone: 518-881-4920<br>
</div>